The Small-Employer ADMT Playbook: What to Do Before January 1

In the first two posts in this series, I covered why Colorado's Automated Decision-Making Technology Act likely reaches your applicant knockout questions, and why it applies to pay, promotion, discipline, and termination decisions, not only hiring.

This post is the plan (subject to change based on the outcome of the Attorney General's guidelines expected later this year). The law takes effect January 1, 2027. There is no small-business exemption. The original 2024 law had one, and the replacement dropped it. Violations are treated as deceptive trade practices under the Colorado Consumer Protection Act, with civil penalties of up to $20,000 per violation. Until 2030, the Attorney General must give you 60 days to cure a violation before enforcing, but that protection doesn't apply to knowing or repeated violations.

That leaves about eleven weeks. Here is how a small or mid-sized employer can use them.

Phase 1: Find Your Tools (Now Through October)

Inventory every system that touches a people decision. Start with the obvious ones: your applicant tracking system, job boards, and any pre-hire assessments. Then look further: your HRIS, compensation platform, performance management system, scheduling and attendance tools, and anything your PEO provides. Ask each department head what software they use to evaluate people. Hiring managers often add tools HR never approved.

For each tool, answer four questions:

  1. What decision does it feed? Hiring, promotion, pay, discipline, or termination.
  2. What does it produce? A score, ranking, classification, recommendation, or pass/fail threshold.
  3. Does a person see that output before the decision is made?
  4. Does it affect Colorado applicants or employees?

Sort the list. Tools that only summarize, organize, translate, draft, route, or present information for a person to review are excluded. So are calculators, databases, and spreadsheets that don't use machine learning and still require human analysis. An AI note-taker or a tool that drafts job descriptions falls on the excluded side. A tool that scores, ranks, or auto-rejects falls on the covered side. If you can't tell, treat it as covered until counsel or the final rules say otherwise.

Clean up your knockout questions. For every automatic screen, confirm the question ties to a bona fide job requirement and complies with existing Colorado law. The Job Application Fairness Act bars asking for age, birth dates, and graduation dates on initial applications. The Chance to Compete Act limits criminal history questions. Remove the knockouts you can't defend. Every screen you drop is one less source of adverse outcomes to explain.

Phase 2: Vendors and Notices (November)

Request developer documentation. The law requires developers to give deployers documentation of a tool's intended uses, known limitations, data categories, and instructions for monitoring and human review. You need it to write accurate notices and explanations. Send each vendor a written request now. Ask whether their system can generate adverse-outcome explanations, log rights requests, and export the records you'll need to keep.

Review your contracts. You're the deployer, and you remain liable for your own decisions. The law voids contract clauses that indemnify a party for its own discriminatory acts. Add terms that require the vendor to deliver documentation, cooperate with data access and correction requests, and notify you of material changes to the tool.

Write your pre-use notice. The law requires clear and conspicuous notice before a covered tool influences a decision. Place it where people will see it: job postings, the application page, and, for employees, wherever you communicate about performance reviews and compensation. A good notice says that an automated tool is used, which decisions it helps make, and how to get more information.

Build your adverse-outcome explanation. Within 30 days of an adverse outcome, the affected person is owed a plain-language explanation. Under the draft rules, it should cover:

Build one template per decision type: screening, interview selection, final hiring, pay, promotion, discipline, and termination. Have employment counsel review the final language.

Phase 3: Human Review, Training, and Records (December)

Designate your human reviewer. People can request meaningful human review of an adverse outcome. The reviewer must be trained, must consider the underlying information rather than defaulting to the tool's output, and must have authority to change the result. The draft rules also expect the reviewer to be independent of the original decision-maker when feasible. In a company without a large HR team, that may be a trained manager outside the reporting line or an outside HR advisor. Decide who it is before January 1.

Set up request intake. Pick one channel, such as a dedicated email address or web form, and log every request. The draft rules set short clocks: rights instructions within 24 hours of a request, acknowledgment of a human review request within 10 days, and a completed review within 45 days. Those timelines may change in the final rules, so build the process now and adjust the dates later.

Train your managers. Under the draft rules, a manager who sees a tool's score or ranking before deciding is presumed to have been influenced by it. Teach managers to document how they reached each decision: interview notes, scorecards, and a written rationale that stands on its own. That record rebuts the presumption and defends you against a discrimination claim.

Keep records for three years. Track which tools you use and why, the notices you posted, every adverse-outcome explanation you sent, and how each rights request was resolved. These records prove compliance, and they are the same records you would need to defend a discrimination charge.

Update your AI use policy and handbook. State which tools are approved for people decisions, who owns them, and the rule that tool outputs inform decisions but don't make them.

Watch the Final Rules

The Attorney General's rule-making hearing is October 26. The final rules will settle open questions, including how "materially influence" is defined, what an adverse-outcome explanation must contain, and possibly how knockout questions are treated. Build your process on the draft rules now, then adjust in December when the final version is published.

Where This Leaves You

Small employers don't need an AI governance department to comply with this law. They need an inventory, a vendor conversation, a few well-written templates, a designated reviewer, and a habit of documenting decisions. Most of that is good HR practice already. The law makes it mandatory.

If you want help working through your tool inventory, vendor contracts, or notices before January 1, that's the work I do with small and mid-sized employers at Your HR Pro.

This post is general information, not legal advice. Have employment counsel review how the law applies to your tools and processes.